https://ir.lib.ncu.edu.tw/handle/987654321/85 Search the Channel s https://ir.lib.ncu.edu.tw/simple-search https://ir.lib.ncu.edu.tw/handle/987654321/111027 title: 以 TOE 架構探討傳統製造業中小企業採納超融合基礎架構之關鍵因素;Key Factors Influencing Hyper-Converged Infrastructure Adoption in Traditional Manufacturing SMEs: A TOE-Based Modified Delphi and AHP Study abstract: 台灣傳統製造業中小企業在數位轉型過程中，逐漸需要以更穩定、可擴充且易於管理的 IT 基礎架構支撐 ERP（Enterprise Resource Planning）、MES（Manufacturing Execution System）、PLC（Programmable Logic Controller）、AS/RS（Automated Storage and Retrieval System）與資料分析等應用。然而，中小企業通常面臨 IT 人力不足、預算有限、舊有系統整合困難與營運中斷風險等限制，使超融合基礎架構（Hyper-Converged Infrastructure， HCI）的採納不僅是技術選型問題，更涉及組織資源、外部支援與風險控管。 既有 HCI 研究多聚焦技術效能、成本比較或大型企業導入經驗，較少從組織層級技術採納觀點探討傳統製造業中小企業如何評估 HCI。本研究以技術、組織與環境（Technology-Organization-Environment， TOE）架構為理論基礎，結合關鍵成功因素觀點，建立傳統製造業中小企業採納 HCI 之初始評估準則。 研究方法上，本研究先透過文獻整理與修正式德爾菲法篩選評估因素，再以層級分析法（Analytic Hierarchy Process， AHP）計算各構面與準則之相對權重。研究資料來自 12位具資訊基礎架構、製造業數位轉型、系統整合、資安或技術採納研究經驗之專家意見，並以個案公司導入情境作為研究背景與問題脈絡。 研究結果顯示，傳統製造業中小企業採納 HCI 時，最重視安全性、高階管理支援、財務實力與技術供應商支援；技術與組織因素之重要性高於環境因素，競爭壓力、政府支持與監管要求之影響相對有限。據此，本研究提出「風險可控導向之 HCI 採納評估框架」，指出中小企業採納 HCI 並非單純追求技術先進性，而是透過資安、管理承諾、財務承受能力與外部技術支援降低導入風險，形成可供同類型企業參考之 IT 基礎架構升級決策邏輯。 關鍵詞：超融合基礎架構、TOE 架構、修正式德爾菲法、層級分析法、中小企業、技術採納;Traditional manufacturing SMEs in Taiwan increasingly require stable， scalable， and manageable IT infrastructure to support ERP， MES， PLC， AS/RS， and data-driven applications. However， SMEs often face limited IT manpower， constrained budgets， legacy system integration issues， and operational continuity risks. Therefore， adopting hyper-converged infrastructure (HCI) is not merely a technical choice， but also an organizational and risk-management decision. Existing HCI studies mainly focus on technical performance， cost comparisons， or large-enterprise implementation cases. Relatively few studies examine HCI adoption among traditional manufacturing SMEs from an organizational technology-adoption perspective. Based on the Technology-Organization-Environment (TOE) framework and the critical success factor perspective， this study develops an evaluation structure for HCI adoption in traditional manufacturing SMEs. Methodologically， this study first identifies initial factors through the literature and then applies the modified Delphi method to screen the evaluation criteria. The Analytic Hierarchy Process (AHP) is subsequently used to calculate the relative weights of the dimensions and criteria. Data were collected from 12 experts with experience in IT infrastructure， manufacturing digital transformation， system integration， cybersecurity， or technology adoption research， while the J Company implementation context serves as the practical background of the study. The results show that security， top management support， financial capability， and technical vendor support are the most important factors influencing HCI adoption. Technology and organizational factors are more influential than environmental factors， while competitive pressure， government support， and regulatory requirements are relatively less influential. Based on these findings， this study proposes a risk-controllability-oriented HCI adoption framework， suggesting that SMEs adopt HCI not simply to pursue technological advancement， but to reduce implementation risks through security， managerial commitment， financial readiness， and external technical support. Keywords: Hyper-Converged Infrastructure， TOE Framework， Modified Delphi Method， Analytic Hierarchy Process， SMEs， Technology Adoption <br> Tue, 20 Oct 2026 08:44:48 GMT https://ir.lib.ncu.edu.tw/handle/987654321/111023 title: 應用敏捷發展法來建立AI服務動態定價模式：以某企業的智慧城市服務為例;應用敏捷發展法來建立AI服務動態定價模式：以某企業的智慧城市服務為例 abstract: 隨著生成式人工智慧技術快速發展，智慧城市服務逐漸由單一系統建置轉向平台化與整合式服務。然而，生成式人工智慧一站式賦能平台在推動過程中，常需同時整合多雲基礎資源、模型服務、資料處理、API 串接、應用模組與維運支援，使成本結構呈現高度複雜與動態變化。若平台服務提供者缺乏清楚的成本治理與報價邏輯，將不易掌握實際成本與毛利，也可能影響客戶對方案價值、預算與導入效益之判斷。 本研究以某企業智慧城市事業單位之生成式人工智慧一站式賦能平台為個案，採用質性個案研究法，並輔以主要公有雲服務供應商之虛擬機器公開定價資料進行多雲成本比較。研究目的在於探討平台服務提供者如何在多雲、多模型與多服務疊加之情境下，建立可被理解、可被追蹤且可持續調整之定價管理機制。本研究結合成本治理、分層訂閱、參數化定價與敏捷檢視與調適精神，提出「成本治理導向之敏捷分層定價框架」。 研究結果顯示，生成式人工智慧平台之定價問題並非單純制定價格表，而是一項連結成本可視性、方案價值表達、報價參數與滾動檢討之持續性管理活動。本研究將平台成本區分為基礎雲端層、模型服務層、應用整合層與治理支援層，並提出基礎導入型、進階應用型與治理整合型三種方案情境，使複雜平台服務得以轉化為客戶較易理解之方案結構。同時，本研究以 Billing Cycle 作為定價衝刺週期，使平台價格能依據成本、用量與客戶回饋進行滾動調整。整體而言，本研究可作為生成式人工智慧平台服務提供者進行成本治理、分層訂閱設計與動態定價管理之參考。;With the rapid development of generative artificial intelligence， smart city services are gradually shifting from single-system implementation toward platform-based and integrated services. However， in the process of promoting a generative AI one-stop enablement platform， service providers often need to integrate multi-cloud infrastructure， model services， data processing， API connections， application modules， and operation and maintenance support. As a result， the cost structure becomes highly complex and dynamic. Without clear cost governance and pricing logic， platform service providers may find it difficult to control actual costs and profit margins， and customers may also face difficulties in evaluating service value， budget requirements， and implementation benefits. This study takes the generative AI one-stop enablement platform of a smart city business unit in a company as the research case. A qualitative case study approach is adopted， supplemented by a comparison of public virtual machine pricing data from major public cloud service providers. The purpose of this study is to explore how platform service providers can establish a pricing management mechanism that is understandable， traceable， and continuously adjustable under the context of multi-cloud， multi-model， and multi-service integration. By integrating cost governance， tiered subscription， parametric pricing， and agile inspect-and-adapt concepts， this study proposes a Cost-Governed Agile Tiered Pricing Framework. The findings indicate that pricing for generative AI platforms is not merely about setting a price list， but rather a continuous management activity that connects cost visibility， value communication， pricing parameters， and rolling review mechanisms. This study classifies platform costs into four layers: cloud infrastructure， model services， application integration， and governance support. It also proposes three service scenarios: basic adoption， advanced application， and governance integration， thereby transforming complex platform services into a more understandable tiered structure for customers. In addition， this study uses the Billing Cycle as a pricing sprint cycle， enabling platform prices to be adjusted continuously based on costs， usage， and customer feedback. Overall， this study provides a reference for generative AI platform service providers in cost governance， tiered subscription design， and dynamic pricing management. <br> Tue, 20 Oct 2026 08:44:33 GMT https://ir.lib.ncu.edu.tw/handle/987654321/111018 title: 基於模型上下文協議 (MCP) 之生成式 AI 自動化防禦架構研究：整合 Wazuh 與 OPNsense 之實作;Research on Generative AI Automated Defense Architecture Based on Model Context Protocol (MCP): Implementation by Wazuh and OPNsense abstract: 隨著企業數位轉型加速，網路攻擊的頻率與複雜度驟增。傳統安全資訊與事件管理（SIEM）系統面臨告警疲勞、回應延遲與未知攻擊盲區等嚴峻挑戰。儘管近年資安領域嘗試導入大語言模型（LLM），但現有架構多受限於系統靜態耦合、依賴輕量模型導致分析效能不佳，且流程往往止步於報告生成，缺乏與底層設備連動的主動防禦能力。為解決上述缺點，本研究提出一套基於模型上下文協議（MCP）之生成式AI（Generative artificial intelligence， GenAI）自動化防禦架構。本架構以Wazuh平台匯集日誌，導入n8n作為MCP伺服器與編排引擎，解決傳統AI代理需靜態嵌入程式碼的限制，實現標準化與靈活配置。在防禦機制上，系統以GenAI（Claude、GPT、Gemini）透過多數決機制為核心決策中樞。接收日誌並正規化後，GenAI調用威脅情資（MISP、AlienVault、OTX）驗證來源IP；若判定該IP為非已知威脅，則進一步對日誌所提供的有效載荷(Payload)進行深層邏輯解碼與語意推理，以識別攻擊行為。一旦Payload判定高風險，系統即觸發n8n編排引擎連動OPNsense防火牆進行即時IP封鎖，形成分鐘級響應的主動防禦閉環。實驗結果顯示，本架構成功透過高階語義推理，識破結合混淆手法的Log4Shell及雙重URL編碼的目錄遍歷等惡意攻擊。相較傳統依賴人工介入與工單等待的流程，本系統將平均修復時間（MTTR）從數小時大幅縮短至數分鐘。本研究不僅實證了MCP協議在標準化AI代理與打破系統靜態耦合上的優勢，更為具備高擴展性與主動阻斷能力的新一代AI防禦，提供了一套具體的實踐準則。;Rapid digital transformation has escalated cyberattack complexity， exposing traditional SIEM systems to alert fatigue and blind spots. Current LLM-driven defenses are often limited by static system coupling， lightweight models， and a lack of active firewall intervention. To address this， we propose an automated Generative AI (GenAI) defense architecture based on the Model Context Protocol (MCP). Using Wazuh for log aggregation， the framework innovatively deploys n8n as an MCP server and orchestrator， breaking the limitations of statically embedded AI agents to enable flexible configuration. GenAI (Claude， GPT， Gemini) serves as the core decision hub. It first queries threat intelligence (MISP， AlienVault， OTX) to verify source IPs; if the IP is benign or unknown， it conducts deep semantic reasoning on packet payloads to identify hidden malicious behaviors. Upon detecting high risks， n8n instantly triggers OPNsense firewalls for real-time IP blocking， forming a minute-level active defense loop. Experiments confirm the successful detection of complex threats， including obfuscated Log4Shell and double URL-encoded directory traversals. By bypassing manual intervention， Mean Time to Repair (MTTR) is significantly reduced from hours to minutes. This study validates MCP′s critical advantage in standardizing AI agents and decoupling systems， providing a highly scalable， proactive framework for next-generation automated AI defense. <br> Tue, 20 Oct 2026 08:44:15 GMT https://ir.lib.ncu.edu.tw/handle/987654321/111014 title: 基於多層特徵動態融合與AI解釋代理之釣魚URL偵測系統;Multi-Layer Dynamic Feature Fusion with AI Explanation Agent for Phishing URL Detection abstract: 釣魚攻擊（Phishing Attack）藉由偽造合法網站統一資源定位符（URL）誘導使用者洩漏帳號密碼與財務資訊，防禦攻擊的核心困境在於黑名單更新速度長期落後於攻擊者輪替網域的速度，使以靜態特徵拼接為主的現有偵測方法在灰色邊界案例中存在系統性誤判，且多數系統缺乏可供資安人員快速複核的決策依據。本研究設計並實作 GateNet，採六層推理架構，核心為 Softmax 正規化閘控多層感知機（Gate MLP），針對每一條輸入 URL 個別計算三路特徵權重，對統計特徵（Stat，41 維）、字元卷積神經網路特徵（Char-CNN，128 維）及語意閘控循環單元特徵（Semantic GRU，128 維）執行融合加權，閘控輸出直接作為可解釋的特徵重要性指標。架構中額外設計領域感知校準層（Step 2a），透過 107 筆可信域名白名單與三層 Typosquatting 偵測機制（可疑 TLD、仿冒前綴、Levenshtein 編輯距離 ≤ 2）雙向校準模型輸出：對仿冒域名調高其釣魚機率以修正漏判，對可信域名則調低機率以避免誤判。並以閘控值審視，調整日誌與 Gemini 2.5 Flash 自然語言威脅報告構成三層可解釋輸出鏈；針對帶路徑良性 URL 的系統性誤判，另提出定向偏差修正方法，透過閘控分布日誌診斷訓練偏差並以針對性資料增強修正。於 587，638 筆修正資料集上，GateNet 測試集準確率達 99.82%、F1 分數 0.9940；逐階段效能自單分支（96.4%–99.7%）、靜態融合（99.79%）演進至動態閘控（99.82%），其改善經 McNemar 檢定具統計顯著性（p=0.003）。在 35 筆人工定向設計的模糊測試集（Fuzzy Test）上，涵蓋合法知名域名、仿冒域名、可疑灰色地帶與新創域名四類邊界情境，屬案例驗證性質，領域感知層將通過數由 20 提升至 32，修正全部 12 筆具可信域名錨點的漏判；其餘 3 筆無可信域名錨點的純通用詞釣魚 URL 為當前架構的結構性限制，完整案例清單與逐筆結果詳見附錄B。;Phishing attacks fabricate the Uniform Resource Locators (URLs) of legitimate websites to trick users into disclosing credentials and financial information. Their core difficulty is that blacklist updates persistently lag behind attackers′ domain rotation， causing detection methods based on static feature concatenation to misclassify gray-boundary cases， while most systems provide no decision rationale that security analysts can quickly review. This study designs and implements GateNet， a six-tier inference architecture whose core is a Softmax-normalized gated multilayer perceptron (Gate MLP). For each input URL it individually computes the weights of three feature streams (per-URL adaptivity) and performs weighted fusion over statistical features (Stat， 41-dim)， character-level convolutional features (Char-CNN， 128-dim)， and semantic recurrent features (Semantic GRU， 128-dim); the gate outputs serve directly as interpretable feature-importance indicators. A domain-aware calibration layer (Step 2a)， comprising a 107-entry brand whitelist and three-tier Typosquatting detection (suspicious TLD， spoofing prefix， Levenshtein distance ≤ 2)， bidirectionally calibrates the output: raising the phishing probability of spoofed domains to correct false negatives， and lowering it for legitimate brand domains to widen the safety margin. A three-tier interpretable output chain—gate inspection， adjustment log， and a Gemini 2.5 Flash threat report—is further provided， together with a targeted bias-correction method that diagnoses training bias from gate-distribution logs and applies targeted data augmentation. On a corrected dataset of 587，638 URLs， GateNet attains 99.82% test accuracy and an F1 score of 0.9940; accuracy progresses from single-branch models (96.4%–99.7%) through static fusion (99.79%) to dynamic gating (99.82%)， an improvement that is statistically significant under McNemar′s test (p=0.003). On a fuzzy-test set of 35 manually curated boundary cases spanning four scenarios (well-known legitimate domains， typosquatted domains， suspicious gray-zone URLs， and new/unknown domains)， the domain-aware layer raises the pass count from 20 to 32， correcting all 12 brand-anchored false negatives; the 3 remaining generic-word phishing URLs without brand anchors represent a structural limitation of the current architecture (the complete case list is provided in Appendix B). <br> Tue, 20 Oct 2026 08:44:10 GMT